AI OBSERVATORY

AI bubble / sustainability tracker

Are end users continuing to pay? Can that revenue support the next round of compute investment?

Not assessed

Current cycle: unconfirmed

Commercial scale is disclosed, but comparable revenue, paid-workload and rental histories are incomplete. No boom or recession classification yet.

2026-09-13Research review dateReviewed disclosures · not a live feed
01Expansion
02Slowing
03Contraction
04Recovery

PressureWatchHealthyNot assessedGrey means insufficient evidence, not low risk.

01Not assessed

Consumer willingness to pay

Retention pending

Track paying users, net ARPU and renewals for ChatGPT, Claude and Gemini.

What changes the light?

Green: paid revenue and retention improve. Amber: slower growth or weaker retention. Red: sustained contraction in both.

02Not assessed

Enterprise adoption

Usage series pending

Compare seats, paid API usage and customer expansion across model providers and Copilot.

What changes the light?

Green: paid workload and revenue expand together. Amber: divergence. Red: both contract; NRR below 100% is a warning.

03Not assessed

Cloud earnings delivery

Forecasts not yet frozen

AWS · Azure · Google Cloud · OCI. Compare results with forecasts set before earnings.

What changes the light?

Green: over 60% of fixed-weight coverage beats forecasts. Amber: 40–60%. Red: below 40%. Show coverage and check paid usage separately.

04Not assessed

GPU rental balance

Matched history pending

Compare the same GPU, region and contract. Falling prices alone do not prove weaker AI demand.

What changes the light?

Green: paid occupancy and renewal economics hold. Amber: one weakens. Red: both deteriorate persistently.

05Not assessed

Cash supports investment

Coverage pending

Track hyperscaler capex coverage and neocloud debt service; separate customer cash from financing.

What changes the light?

Green: adjusted operating cash covers capex and debt obligations are covered. Amber: a funding gap with available liquidity. Red: a cash shortfall or debt-service failure.

Verified disclosures, not a cycle verdict

OpenAI

$2B / month

Revenue at the March 31, 2026 disclosure; $24B annualized, not realized annual revenue or contracted ARR. Enterprise represented over 40%.

Official disclosure

Anthropic

> $47B run-rate

The May 28, 2026 announcement reported over $47B revenue run-rate earlier in May, not recognized annual revenue. Comparable quarters, retention and cash burn remain to be added.

Official disclosure

What to check next

  1. Does revenue growth come with renewals, customer expansion and paid usage?
  2. Do cloud earnings beats reflect AI services, hardware sales, resale or recognition timing?
  3. Can GPU renewal cash cover operations, debt service and equipment replacement?
How the cycle and lights are assessed

For seasonally adjusted, same-sample end-user revenue and fixed-quality paid workload, calculate quarterly growth g and its change Δg. Both series must agree for two quarters: positive g and Δg = expansion; positive g and negative Δg = slowing; both negative = contraction; negative g and positive Δg = recovery. Near-zero, conflicting, stale or missing data remain unclassified.

Lights assess individual evidence, do not map one-to-one to cycle stages, and are not averaged into a bubble score. Green does not mean cheap valuation. Financing is not revenue; model-company and cloud revenue must not be double-counted. The 60%/40% earnings-breadth bands are initial, unbacktested rules.

Review cadence: weekly rental quotes, monthly paid usage and run-rate, quarterly earnings and cash flow. This page is a manual research snapshot; automatic updates are not connected and earnings forecasts have not yet been frozen.